Thanks greg, that formula was exactly what I was looking for. Except now when I run it on my data I get the following error:
"Error in model.matrix.default(mt, mf, contrasts) : cannot allocate vector of length 2043479998" I know there are probably many 2-way interactions that are zero so I thought I could save space by removing these. Is there some way that can just delete all the two way interactions that are zero and keep the columns that have non-zero entries? I think that will significantly cut down the memory needed. Or is there just another way to get around this? thanks, Matt On Tue, Mar 1, 2011 at 3:56 PM, Greg Snow <greg.s...@imail.org> wrote: > You can use ^2 to get all 2 way interactions and ^3 to get all 3 way > interactions, e.g.: > > lm(Sepal.Width ~ (. - Sepal.Length)^2, data=iris) > > The lm.fit function is what actually does the fitting, so you could go > directly there, but then you lose the benefits of using . and ^. The Matrix > package has ways of dealing with sparse matricies, but I don't know if that > would help here or not. > > You could also just create x'x and x'y matricies directly since the variables > are 0/1 then use solve. A lot depends on what you are doing and what > questions you are trying to answer. > > -- > Gregory (Greg) L. Snow Ph.D. > Statistical Data Center > Intermountain Healthcare > greg.s...@imail.org > 801.408.8111 > > >> -----Original Message----- >> From: Matthew Douglas [mailto:matt.dougla...@gmail.com] >> Sent: Tuesday, March 01, 2011 1:09 PM >> To: Greg Snow >> Cc: r-help@r-project.org >> Subject: Re: [R] Regression with many independent variables >> >> Hi Greg, >> >> Thanks for the help, it works perfectly. To answer your question, >> there are 339 independent variables but only 10 will be used at one >> time . So at any given line of the data set there will be 10 non zero >> entries for the independent variables and the rest will be zeros. >> >> One more question: >> >> 1. I still want to find a way to look at the interactions of the >> independent variables. >> >> the regression would look like this: >> >> y = b12*X1X2 + b23*X2X3 +...+ bk-1k*Xk-1Xk >> >> so I think the regression in R would look like this: >> >> lm(MARGIN, P235:P236+P236:P237+....,weights = Poss, data = adj0708), >> >> my problem is that since I have technically 339 independent variables, >> when I do this regression I would have 339 Choose 2 = approx 57000 >> independent variables (a vast majority will be 0s though) so I dont >> want to have to write all of these out. Is there a way to do this >> quickly in R? >> >> Also just a curious question that I cant seem to find to online: >> is there a more efficient model other than lm() that is better for >> very sparse data sets like mine? >> >> Thanks, >> Matt >> >> >> On Mon, Feb 28, 2011 at 4:30 PM, Greg Snow <greg.s...@imail.org> wrote: >> > Don't put the name of the dataset in the formula, use the data >> argument to lm to provide that. A single period (".") on the right >> hand side of the formula will represent all the columns in the data set >> that are not on the left hand side (you can then use "-" to remove any >> other columns that you don't want included on the RHS). >> > >> > For example: >> > >> >> lm(Sepal.Width ~ . - Sepal.Length, data=iris) >> > >> > Call: >> > lm(formula = Sepal.Width ~ . - Sepal.Length, data = iris) >> > >> > Coefficients: >> > (Intercept) Petal.Length Petal.Width >> Speciesversicolor >> > 3.0485 0.1547 0.6234 - >> 1.7641 >> > Speciesvirginica >> > -2.1964 >> > >> > >> > But, are you sure that a regression model with 339 predictors will be >> meaningful? >> > >> > -- >> > Gregory (Greg) L. Snow Ph.D. >> > Statistical Data Center >> > Intermountain Healthcare >> > greg.s...@imail.org >> > 801.408.8111 >> > >> > >> >> -----Original Message----- >> >> From: r-help-boun...@r-project.org [mailto:r-help-bounces@r- >> >> project.org] On Behalf Of Matthew Douglas >> >> Sent: Monday, February 28, 2011 1:32 PM >> >> To: r-help@r-project.org >> >> Subject: [R] Regression with many independent variables >> >> >> >> Hi, >> >> >> >> I am trying use lm() on some data, the code works fine but I would >> >> like to use a more efficient way to do this. >> >> >> >> The data looks like this (the data is very sparse with a few 1s, -1s >> >> and the rest 0s): >> >> >> >> > head(adj0708) >> >> MARGIN Poss P235 P247 P703 P218 P430 P489 P83 P307 P337.... >> >> 1 64.28571 29 0 0 0 0 0 0 0 0 0 0 >> >> 0 0 0 >> >> 2 -100.00000 6 0 0 0 0 0 0 0 1 0 0 >> >> 0 0 0 >> >> 3 100.00000 4 0 0 0 0 0 0 0 1 0 0 >> >> 0 0 0 >> >> 4 -33.33333 7 0 0 0 0 0 0 0 0 0 0 >> >> 0 0 0 >> >> 5 200.00000 2 0 0 0 0 0 0 0 0 0 0 >> >> -1 0 0 >> >> 6 -83.33333 12 0 -1 0 0 0 0 0 0 0 0 >> >> 0 0 0 >> >> >> >> adj0708 is actually a 35657x341 data set. Each column after "Poss" >> is >> >> an independent variable, the dependent variable is "MARGIN" and it >> is >> >> weighted by "Poss" >> >> >> >> >> >> The regression is below: >> >> fit.adj0708 <- lm( adj0708$MARGIN~adj0708$P235 + adj0708$P247 + >> >> adj0708$P703 + adj0708$P430 + adj0708$P489 + adj0708$P218 + >> >> adj0708$P605 + adj0708$P337 + .... + >> >> adj0708$P510,weights=adj0708$Poss) >> >> >> >> I have two questions: >> >> >> >> 1. Is there a way to to condense how I write the independent >> variables >> >> in the lm(), instead of having such a long line of code (I have 339 >> >> independent variables to be exact)? >> >> 2. I would like to pair the data to look a regression of the >> >> interactions between two independent variables. I think it would >> look >> >> something like this.... >> >> fit.adj0708 <- lm( adj0708$MARGIN~adj0708$P235:adj0708$P247 + >> >> adj0708$P703:adj0708$P430 + adj0708$P489:adj0708$P218 + >> >> adj0708$P605:adj0708$P337 + ....,weights=adj0708$Poss) >> >> but there will be 339 Choose 2 combinations, so a lot of independent >> >> variables! Is there a more efficient way of writing this code. Is >> >> there a way I can do this? >> >> >> >> Thanks, >> >> Matt >> >> >> >> ______________________________________________ >> >> R-help@r-project.org mailing list >> >> https://stat.ethz.ch/mailman/listinfo/r-help >> >> PLEASE do read the posting guide http://www.R-project.org/posting- >> >> guide.html >> >> and provide commented, minimal, self-contained, reproducible code. >> > > ______________________________________________ R-help@r-project.org mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.